如何将[Batch_Size，A，B]张量乘以[Batch_Size，B]张量以产生[Batch_Size，A]张量？

Question

Question as in title — how do I multiply two tensors of those shapes to yield a tensor of that shape?

[Batch_Size, A, B] x [Batch_Size, B] = [Batch_Size, A]

tf.matmul() requires the tensors be of the same rank.

Answer 1

import tensorflow as tf

tensor1 = tf.zeros([10, 64, 64])
tensor2 = tf.zeros([10, 64])

# dot = tf.matmul(tensor1, tensor2)  # NOT THIS
dot = tf.einsum('ijk,ik->ij', tensor1, tensor2)

with tf.Session() as sess:
    print(sess.run(dot).shape)